Abstract Spatial flow data are often used to represent spatial interaction phenomena such as daily commuting trips, human or animal migrations, and the exchanges of commodities, capital, or even information between regions. With the increasingly available large volume of flow data in fine spatiotemporal resolution, exploratory spatial data analysis (ESDA) has become more important than ever to gain understanding of the data and the story behind it. A major group of flow-related ESDA methods focus on measuring spatial associations, which proves useful in improving the prediction power and interpretability of spatial interaction model (SIM), as well as in identifying local clusters and outliers of flow events. This paper introduces a new spatial statistical method called BiFlowLISA—a local indicator of spatial association of bivariate flow data. BiFlowLISA evaluates the association between two types of flows in close proximity, in other words, how the value of type-I flows associate with the value of nearby type-II flows. We develop BiFlowLISA by extending the local bivariate Moran's I to the flow context. We also put forth its global version to measure the global patterns, and another variant of BiFlowLISA to measure both spatial and in-situ correlations at the same time. Several flow-specific issues are discussed and solved, including flow neighbor definition, OD matrix sparsity, and conditional permutation. We experiment with synthetic datasets to verify its functionality and to summarize its characteristics. A case study of taxi and ride-hailing services in New York City demonstrates its usefulness in the comparative analysis of the spatial patterns of two types of travel flows. More applications of BiFlowLISA await to be explored in the future.

中文翻译：

---

BiFlowLISA：测量双变量流数据的空间关联

摘要 空间流数据常被用来表示空间交互现象，如日常通勤出行、人或动物迁徙，以及区域间商品、资本甚至信息的交换等。随着以精细时空分辨率提供的大量流量数据越来越多，探索性空间数据分析 (ESDA) 对于了解数据及其背后的故事变得比以往任何时候都更加重要。与流相关的 ESDA 方法主要集中在测量空间关联，这在提高空间交互模型 (SIM) 的预测能力和可解释性以及识别流事件的局部集群和异常值方面被证明是有用的。本文介绍了一种新的空间统计方法，称为 BiFlowLISA——双变量流数据空间关联的局部指标。BiFlowLISA 评估靠近的两种类型流之间的关联，换句话说，I 类流的值与附近 II 类流的值如何关联。我们通过将局部二元 Moran's I 扩展到流上下文来开发 BiFlowLISA。我们还提出了它的全局版本来测量全局模式，以及 BiFlowLISA 的另一个变体来同时测量空间和原位相关性。讨论并解决了几个特定于流的问题，包括流邻居定义、OD 矩阵稀疏性和条件排列。我们用合成数据集进行实验以验证其功能并总结其特征。纽约市出租车和叫车服务的案例研究证明了它在比较分析两种出行流的空间模式中的有用性。